Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Chapter 5 The Thermal Evolution of an Earth with Strong Subduction Zones Published in Geophysical Research Letters by C. P. Conrad and B. H. Hager, 26, 3041-3044, 1999. Copyright by the American Geophysical Union. Abstract. It is commonly supposed that plate tectonic rates are controlled by the temperature-dependent viscosity of Earth's deep interior. If this were so, a small decrease in mantle temperature would lead to a large decrease in global heat transport. This negative feedback mechanism would prevent mantle temperatures from changing rapidly with time. We propose alternatively that convection is primarily resisted by the bending of oceanic lithosphere at subduction zones. Because lithospheric strength should not depend strongly on interior mantle temperature, this relationship decreases the sensitivity of heat ow to changes in interior mantle viscosity, and thus permits more rapid temperature changes there. The bending resistance is large enough to limit heat ow rates for eective viscosities of the lithosphere greater than about 10 Pa s, and increases with the cube of plate thickness. As a result, processes that aect plate thickness, such as small-scale convection or subduction initiation, could profoundly inuence Earth's thermal history. 23 175 5.1 Introduction Plate velocities are typically thought to depend on the viscosity of the mantle's deep interior, which depends strongly on temperature. In this view, as the earth cools, mantle viscosity increases, which forces slower convection and less ecient heat transport. This slows the mantle's cooling rate and stabilizes its internal temperature [e.g., Davies, 1980; Tozer, 1972]. Mantle temperatures can change more rapidly with time if convection is inuenced by the lithosphere, which is colder, and therefore stier, than the underlying mantle. In simple convection calculations that include temperaturedependent viscosity, Christensen [1985] found surface heat ow to depend more on the viscosity of the cold thermal boundary layer than on that of the underlying mantle. Because lithospheric viscosity depends on surface temperatures, which are thought to have changed little since the Archean, Christensen's [1985] calculations predict nearly constant heat ow. The decoupling of heat ow from temperature prevents heat loss rates from slowing as the interior cools. This leads to greater present-day cooling rates in thermal evolution models that reproduce current mantle temperatures [Christensen, 1985]. The notion that the viscosity of the lithosphere limits mantle heat ow has received some criticism, particularly because Christensen's [1985] convection models do not produce realistic subduction zones. Instead, these models describe convection beneath a \stagnant lid," where only a small portion of the upper thermal boundary layer participates in downwelling beneath nearly stationary surface plates. Gurnis [1989] produced more realistic subduction and plate motions by including \weak zones" of low viscosity uid at plate boundaries. These weak zones diminish the role of the lithosphere in controlling plate velocities. Because interior mantle viscosity remains as the controlling parameter, mantle temperatures are again stabilized. The geometry of Earth's subduction zones requires cold, strong, lithosphere to deform as it bends and descends into the mantle. Conrad and Hager [1999a] show that this bending deformation provides at least as much resistance to plate motions as viscous deformation of the underlying mantle. Thus, the rheological properties of 176 the bending lithosphere should play an important role in determining plate velocities. If this is the case, the negative feedback mechanism that stabilizes temperatures is disrupted while plate motions and subduction are maintained. In this study, we include the bending resistance in studies of parameterized convection to show how plate bending at subduction zones could play an important role in Earth's thermal evolution. 5.2 Parameterized Convection Models The relationship between convective heat transport and interior temperature is typically investigated in the context of Rayleigh-Bernard convection, where a constant temperature boundary condition is imposed at the base of the system. For this scenario, the dimensionless heat ux is given by the Nusselt number, Nu, which is the ratio of the heat ow due to convection to that due to conduction alone. For steady convection, Nu is often related to the Rayleigh number of the system, Ram , by a power law [e.g., Davies, 1980]: Nu Ram where Ram = gT D3 m (5.1) where , g, , and are the density, gravitational acceleration, thermal expansivity and diusivity, and T is the temperature dierence across the system's depth, D. For a system with constant viscosity m and constant basal temperature, boundary layer theory gives = 1=3 [e.g., Turcotte and Oxburgh, 1967]. For internal heating, both the denition of Nu and the value of predicted by boundary layer theory dier [e.g., O'Connell and Hager, 1980]. The strength of the negative feedback mechanism that stabilizes mantle temperatures depends on . As the earth cools, Ram decreases because m depends on temperature. If > 0, this decreases Nu, and slows the mantle's rate of temperature change. Conversely, knowledge of the mantle's cooling rate allows us to distinguish among convection models that yield dierent values of [e.g., Christensen, 1985]. 177 The earth's present rate of secular cooling is obtained from estimates of the Urey ratio, which is the ratio of current mantle heat production to total heat ow. The latter has been estimated at about 39 10 W, after the contribution from radioactivity in the continental crust, 5 10 W, is subtracted from the worldwide total heat ow [e.g., O'Connell and Hager, 1980; Sclater et al., 1980; Stein, 1995]. The rate of internal heating can be estimated from the abundances of the heat producing elements, U, Th, and K, in the mantle. Their relative abundances are relatively well constrained to a ratio of about 1:4:10000 [e.g., Jochem et al., 1983], but their absolute abundance is more uncertain. Geochemical models of the primitive mantle yield 21 ppb for the present abundance of U, giving a current mantle heat production of 15 10 W [Jochem et al., 1983]. For comparison, if the mantle were solely composed of the material sampled by mid-ocean ridges (MORB), with a concentration of 3 ppb for U, the mantle heat production would be 2:4 10 W [Kellogg et al., 1999]. These estimates, when compared with the mantle heat ow, yield a Urey ratio of 0.4 for models using a primitive mantle and 0.06 for a mantle composed of MORB source. Thus, 60% or more of the current mantle heat ow may represent secular cooling. A thermal history for the earth is obtained by assuming a heating history from estimates of the Urey ratio and integrating heat ow rates from (5:1) backward in time from Earth's present thermal state. For = 1=3, Christensen [1985] nds that the mantle cools from a molten state to its present one in signicantly less than the expected 4.5 billion years if more than about 15% of current mantle heat ow is due to secular cooling. Thus, to allow Urey ratios less than 0.85, the feedback mechanism that stabilizes mantle temperatures must be diminished, which implies a decrease in . If < 0:1, Christensen [1985] obtains plausible thermal histories for Urey ratios between 0.3 and 0.6. These values are consistent with the above estimates of the Urey ratio, but, as described above, Christensen's [1985] convection models that yield < 0:1 do not produce realistic plate motions or subducting slabs. In what follows, we introduce an analysis based on plate bending in subduction zones that produces small while still allowing realistic plate and slab behavior. 12 12 12 12 178 5.3 Plate Bending and Mantle Heat Flow The mantle loses much of its heat through the formation of oceanic lithosphere. The total heat ow from an oceanic plate, expressed here as N , a dimensionless number analogous to a Nusselt number, is given by: v (1=2) 4D p p = N = 2D L hs (5.2) q where vp is the plate velocity, L is the ridge to trench plate length, and hs = 2 L=vp is the plate thickness at the time of subduction [e.g., Turcotte and Schubert, 1982, p. 280-3]. It has been suggested that oceanic plates only thicken for 80 Ma, the age at which seaoor attening is observed to begin [e.g., Sclater et al., 1980]. If additional heat transport for older plates limits their thickness to some maximum value hm < hs, then N should be larger than that given by (5:2). For Earth, the additional heat ow is less than 10%, and is important only for the oldest plates [Sclater et al., 1980]. The velocity of a plate can be estimated by considering the energy budget of a single convecting cell [e.g., Conrad and Hager, 1999a; Richter, 1984; Turcotte and Oxburgh, 1967]. Convection is driven by the potential energy release of a descendp ing slab, given by = gT vplshs = [Conrad and Hager, 1999a] where ls is the eective slab length. Slab reheating is accompanied by cooling of the surrounding mantle, so a slab's net buoyancy change as it descends should be negligible. Thus, ls is determined by the depth of convective circulation, and can be taken as a constant. Potential energy release is typically balanced by viscous dissipation within the shearing mantle, given by Conrad and Hager [1999a] as m = 3m vp (L=D + 2). Conrad and Hager [1999a] also derive the expression l = 2l vp (hs =R) for the viscous dissipation within the bending subducting slab and note that l could be as large as, or even much greater than, m . Here l is the eective viscosity of the lithosphere and R is the radius of curvature of the subducting plate. Setting = m + l pe vd vd 2 2 3 vd vd pe 179 vd vd and solving for vp yields: p gT ls hs = vp = 3m (L=D + 2) + 2l (hs =R)3 (5.3) q We now solve for plate thickness hs = 2 L=vp using (5:3): p L 12 R3 (L=D + 2) p hs = m l ls Ral 8 L=ls !(1=3) (5.4) where we have dened a dimensionless \lithospheric" Rayleigh number: gT R3 Ral = l (5.5) that is a measure of convective instability resisted by the bending lithosphere. p It is clear from (5:4) that as Ral approaches 8 L=ls , the thickness hs becomes innite. This occurs because a plate slowed by lithospheric bending will thicken due to increased cooling, and thus slow even further because a thicker plate is more dicult to bend. This runaway process leads to innitely thick plates, which are not plausible for the earth, so we impose a maximum plate thickness hm. Thus, if hs in (5:4) is larger than hm , we obtain an expression for N by setting hs = hm in (5:3) and applying the resulting expression for vp to (5:2). In terms of Ram and Ral, this yields: ! = 4 ls hm Ram Ral (5.6) N= = LD 3 (L=D + 2) Ral + 2Ram hm =D We now study limiting cases for weak and strong lithosphere. (1 2) 3 2 5.3.1 3 3 Weak Bending Lithosphere If lithosphere is weak or does not bend sharply, Ral is large. If Ral >> 8pL=ls, applying (5:4) to (5:2) yields: 16ls N = Ram 3 L (L=D + 2) 2 180 !(1=3) (5.7) which applies if Ram is suciently large that hs < hm. In this case, convection is adequately described by boundary layer theory, giving 1=3. If, however, Ram is suciently small that hs > hm , plates reach their maximum thickness before subducting and (5:6) yields: 4lshm N = Ram = 3 LD (L=D + 2) !(1=2) (5.8) 3 2 In this scenario, plate velocities are slowed by resistance in the mantle until the plate thickness saturates. Because slab thickness is constant, faster plates are no longer slowed because their slabs are thinner. As a result, N is more sensitive to changes in Ram than it is for simple boundary layer theory, as shown by the power-law exponent of = 1=2. 5.3.2 Strong Bending Lithosphere p A strong resistance to bending occurs if Ral is less than or about 8 L=ls , in which case plates always reach their maximum thickness, so (5:6) applies. If Ram is small so that the denominator of (5:6) is dominated by Ral, then (5:6) reduces to (5:8). Otherwise, if Ram is large, (5:6) becomes: 2l D N = Ral = s Lh 2 3 2 !(1=2) (5.9) 2 m Here the bending dissipation l , overwhelms the dissipation in the underlying mantle m , and thus controls plate velocities. This causes N to be sensitive only to Ral . If Ral is independent of changes in Ram , 0 in (5:1) is implied. vd vd 5.4 Application to Earth We now determine how N varies with Ram for a set of parameter values that characterize the earth. We use m = 3300 kg m , g = 10 m s , = 3 10 K , T = 1200 K, = 10 m s , D = 2500 km, R = 200 km [Bevis, 1986], and 3 6 2 1 181 2 5 1 10 3 23 ηl=10 Pa s L=2500 km h =100 km m N 1/3 β∼ 10 L=5000 km 2 L=7500 km β∼0 1 β∼ 1/ 2 L=10000 km 10 6 10 10 7 8 10 Ram 9 10 10 10 Figure 5.1: A log-log plot of N vs. Ram for four dierent plate lengths, L, using the \preferred" values of l = 10 Pa s and hm = 100 km. The slopes, , are the exponent of the N Ram relation. 23 ls = 1000 km, and assume that these values remain constant throughout Earth history. In doing so, we assume that the dynamical eect of any change in T , by at most 50%, will be overwhelmed by the accompanying orders of magnitude change in mantle viscosity. Because these temperature-induced changes in m should have the largest eect on the mantle Rayleigh number, we vary m to generate a range of values for Ram. The maximum plate thickness and the lithosphere viscosity are somewhat dicult to estimate and may have been dierent in the past. To see how heat transport is aected by each of these quantities, we plot N as a function of Ram for various combinations of hm and l (Figures 5.1 to 5.3), but show results for \preferred" values in Figure 5.1. We estimate hm = 100 km by assuming plates achieve their maximum thickness after 80 Ma. The eective viscosity of the bending lithosphere is likely the result not only of viscous ow, but also of plastic and brittle deformation associated with non-Newtonian creep and faulting. Conrad and Hager [1999a] estimate this eective viscosity to be of order l = 10 Pa s, a value a few orders of magnitude 23 182 stier than estimates for the mantle. We vary L between 2500 and 10000 km, the approximate range of plate lengths on Earth. To calculate N , we rst use (5:4) to calculate hs . If hs > hm, we use (5:6) to calculate N , otherwise, we use (5:2). The functional dependence of N on Ram varies dierently for dierent plate lengths, L. For small Ram , the N versus Ram curve has a slope of 1=2 (Figure 5.1), as predicted by (5:8). For large Ram , long plates show a slope of 0 while short plates give 1=3 (Figure 5.1). The divergence between these two behaviors occurs because short plates do not thicken suciently for the bending resistance to become large. Long plates, on the other hand, do have time to thicken, so their bending resistance becomes large enough to limit plate velocities if Ral < 8pL=ls. Thus, for weak lithosphere leading to large Ral (Figure 5.2a), plates of all lengths are governed by 1=3, as predicted by (5:7) for weak bending resistance. If the lithosphere is strong (Figure 5.2b), plates of all lengths are dominated by the bending resistance, which yields 0, as predicted by (5:9). Intermediate behavior occurs for l = 10 Pa s, for which only long plates show 0 (Figure 5.1). If the mantle cools, the negative buoyancy of the lithosphere decreases (assuming constant surface temperatures). As a result, the lithosphere should become less unstable; if small-scale convection erodes the lithospheric base, it should become less eective as the mantle cools, increasing hm. Thicker plates produce smaller heat ow, so larger hm should lead to smaller N (compare N for long plates in Figures 5.1 and 5.3). Thus, if hm increases with decreasing Ram, must be greater than zero. By comparing the values of N for hm = 50 and 100 km (Figures 5.3a and 5.1), we estimate that if hm were to double due to a decrease of Ram by two orders of magnitude, we would obtain 0:15. Such a large increase in hm seems unlikely, so should probably be somewhat less than 0:15. Cooler mantle temperatures produce larger mantle viscosities, but may also increase the eective viscosity of the lithosphere. If bending stresses control convection, N decreases with increasing l , leading to > 0 (compare N for long plates in Figures 5.1 and 5.2b). An increase in l could also cause bending stresses to suddenly p become important if Ral becomes smaller than 8 L=ls (compare curves for long 23 183 10 3 22 a) ηl=10 Pa s h =100 km m 10 2 Plate Lengths: L=2500 km L=5000 km L=7500 km L=10000 km β∼ 1/ 2 N 1/3 β∼ 1 10 3 10 b) ηl=1024 Pa s N h =100 km m 10 Plate Lengths: L=2500 km L=5000 km L=7500 km L=10000 km 2 β∼0 1 10 6 10 10 7 10 Ra 8 10 9 10 10 m Figure 5.2: N vs. Ram curves for hm = 100 km and (a) l = 10 Pa s or (b) l = 10 Pa s. To examine a full range of lithospheric viscosities, l , compare to Figure 5.1. 22 24 184 3 10 23 a) ηl=10 Pa s L=2500 km h =50 km N m 1/3 β∼ L=5000 km L=7500 km 2 10 β∼0 β∼ 1/ 2 L=10000 km 1 10 3 10 b) ηl=1023 Pa s L=2500 km h =200 km m N 1/3 β∼ L=5000 km 2 10 L=7500 km L=10000 km β∼0 1 10 6 10 10 7 10 Ra 8 9 10 10 10 m Figure 5.3: N vs. Ram curves for l = 10 Pa s and (a) hm = 50 km or (b) hm = 200 km. To examine a full range of plate thicknesses, hm, compare to Figure 5.1. 23 185 plates in Figures 5.2a and 5.1). Thus, mantle cooling could cause bending slabs to become more eective at resisting convection than the shearing mantle, causing to drop from 1=3 to 0 at some time. 5.5 Conclusions We have shown that if plate bending in subduction zones is important in controlling plate velocities, heat ow is less sensitive to interior viscosities and temperatures, so more rapid changes in mantle temperature are permitted. Thus, plate bending could be a mechanism by which values of less than 0.1 can be obtained in the relationship Nu Ra . As discussed above and by Christensen [1985], must be small to reconcile parameterized convection models with the observed secular cooling of the mantle, expressed by estimates of Urey ratios near 0.4. The plate bending mechanism is preferable to other models of convection that yield < 0:1 because it arises naturally from subduction zone geometry, and thus produces realistic plate and slab behavior. We conclude that small values of are only achieved if bending subducting slabs have an eective viscosity of order 10 Pa s or more. For viscosities close to this value, only an earth with large, Pacic-sized plates that can grow suciently thick before they subduct will experience a decrease in . Because the plate thickness is so inuential in determining plate velocities, any process that aects plate thickness could be an essential aspect of plate tectonics, and could greatly inuence Earth's thermal evolution. Such processes may include small-scale convection beneath plates, which may limit the plate thickness through basal erosion, or the physical details of subduction initiation, which may control how large, and thus how thick, plates can become. 23 Acknowledgments. This work was supported by National Science Foundation grant 9506427-EAR and by a National Science Foundation Graduate Research Fellowship. We thank M. Kameyama and an anonymous referee for helpful reviews and P. Molnar for insightful comments. 186